A Framework for Medical and Stock Image Retrieval in Wavelet Domain using Color, BDIPs and BVLC in Covariance
In the present scenario, content based image retrieval (CBIR) for diverse image collection is fascinating and essential task. Many researchers have worked and presented diverse feature descriptors. However, most of them are limited in discriminative power. In this work, we increased the retrieval rate of CBIR by integrating the multi resolution domain based color, BDIP (Block difference of inverse probabilities) and BVLC (Block variation of local correlation coefficients) in mean and covariance matrix. The multi resolution domain has been exploited using the Discrete Wavelet Transform (DWT) on R, G and B images respectively. The coefficients of DWT are used to estimate the proposed integrated feature vector. The BDIP computes edges and valleys details using local intensity maxima and minima respectively and BVLC computes texture smoothness details using the differences between local correlation coefficients The presented approach is tested on six benchmark databases namely Corel 1k, 5k and 10k, Holiday, Caltech 101 and Histopathology image databases. The proposed integration of multiresolution feature descriptor is compared with previous approaches and results in terms of precision, recall and G-measure are clearly shown that proposed approach is superior in performance for diverse image collections.